Artificial Intelligence

Legal Framework

Artificial intelligence systems fall within the scope of the AI Act, which defines AI systems by their capacity to infer — a capability that distinguishes them from traditional software. Recital 12 establishes that the definition turns on functional characteristics rather than labels:

"A key characteristic of AI systems is their capability to infer. This capability to infer refers to the process of obtaining the outputs, such as predictions, content, recommendations, or decisions, which can influence physical and virtual environments"
— AI Act Recital 12

Where an AI system constitutes a safety component embedded in a product covered by Union harmonisation legislation, the product manufacturer assumes the obligations of a provider under the AI Act (Recital 87). General-purpose AI systems — those capable of serving a variety of purposes through integration of a general-purpose AI model — are addressed under Recital 100. High-risk AI systems additionally require cybersecurity protections proportionate to their risk profile, as Recital 76 mandates:

"Cyberattacks against AI systems can leverage AI specific assets, such as training data sets (e.g. data poisoning) or trained models (e.g. adversarial attacks or membership inference), or exploit vulnerabilities in the AI system's digital assets or the underlying ICT infrastructure."
— AI Act Recital 76

These provisions operate alongside the GDPR, which continues to govern the lawful basis for processing personal data used in training, testing, and deploying AI systems.

Key Developments

The enforcement architecture for AI systems remains under active construction. The EDPB and EDPS have advocated for a coordinated supervisory approach, recommending that the EDPB play an advisory role on data protection aspects within AI regulatory sandboxes. The AI Office holds exclusive competence for supervising compliance of AI systems based on general-purpose AI models when the model and system share the same provider (EDPB-EDPS Joint Opinion 1/2026, §30). The Proposal under discussion extends this competence to AI systems integrated into designated very large online platforms or search engines under the DSA.

National DPAs retain authority over data protection dimensions of AI processing. The EDPB has emphasized that where a national DPA is actively involved in supervising an AI system developed in an EU-level sandbox, consistency mechanisms must ensure coherent application across jurisdictions.

Status of the Debate

This topic is actively contested and regulator-defined. The AI Act's framework is newly in force, and its interaction with GDPR requirements — particularly around lawful basis for training data, automated decision-making under Article 22 GDPR, and the allocation of controller versus provider responsibilities — remains unresolved. The EDPB-EDPS Joint Opinion reveals ongoing institutional negotiations over supervisory competence between the AI Office, national DPAs, and the EDPS. No court has yet produced a definitive ruling on the boundaries between AI Act obligations and GDPR data protection requirements for AI systems. What would resolve the open questions: a CJEU preliminary reference clarifying whether AI Act conformity assessments satisfy or complement GDPR accountability obligations, and a settled EDPB guideline on appropriate lawful bases for AI training datasets.

Practical Guidance

• Classify your AI system accurately. Determine whether your system meets the AI Act definition by assessing its inferential capability — systems based solely on rules defined by natural persons fall outside the scope (Recital 12).
• Identify provider obligations for embedded AI. Where high-risk AI is integrated into a regulated product, the product manufacturer — not the AI developer — may bear provider-level compliance duties (Recital 87).
• Implement AI-specific cybersecurity measures. Address training data integrity (data poisoning), model robustness (adversarial attacks), and infrastructure vulnerabilities as distinct risk vectors (Recital 76).
• Map supervisory competence before deployment. Determine whether your AI system falls under the AI Office's exclusive competence (general-purpose AI models, VLOP/VLOSE integration) or national DPA jurisdiction, and engage the relevant authority early.
• Maintain GDPR compliance independently. The AI Act does not displace GDPR obligations; secure a valid lawful basis for all personal data processed in training, testing, and inference phases.